143174 research outputs found
Sort by
PACESS: Practical AI-based Cell Extraction and Spatial Statistics for large 3D bone marrow tissue images
No full textAlthough the molecular regulation of hematopoiesis is well characterized, the spatial organization of hematopoietic cells within bone marrow (BM) remains unclear. Advances in microscopy have produced increasingly detailed images of murine BM, yet accurate and scalable methods to extract and analyze these complex datasets are limited. The high cellular density of the BM complicates image segmentation, and current spatial analyses are often restricted to pairwise comparisons, unsuitable for investigating interactions between more than two cell types simultaneously. To overcome these limitations, we developed PACESS, a readily applicable neural network-based framework that classifies hundreds of thousands of cells in 3D BM samples and applies spatial statistical methods to evaluate multicellular interactions. Using PACESS, we investigate the spatial organization of T cells, megakaryocytes and leukemic cells, revealing that distinct leukemic clusters generate diverse, previously unrecognized neighborhood within the same BM cavity. PACESS thus provides a powerful tool to dissect BM architecture
Reinforcement learning increases wind farm power production by enabling closed-loop collaborative control
No full textTraditional wind farm control operates each turbine independently to maximize individual power output. However, coordinated wake steering across the entire farm can substantially increase the combined wind farm energy production. Although dynamic closed-loop control has proven effective in flow control applications, wind farm optimization has relied primarily on static, low-fidelity simulators that do not resolve critical dynamic turbulent fluctuations in the flow. In this work, we present a reinforcement learning (RL) controller trained using high-fidelity large-eddy simulation (LES), enabling real-time response to atmospheric turbulence through collaborative,
dynamic control strategies. In a three wind turbine test case, our RL controller achieves a 4.30% (95% CI = [4.10%, 4.49%]) increase in wind farm power output compared to baseline operation, nearly doubling the 2.19% (95% CI = [1.98%, 2.39%]) gain from static optimal yaw control and
a substantial increase over the gain from global wind direction based dynamic control obtained through Bayesian optimization of 2.67% (95% CI = [2.47%, 2.87%]). These results establish that reinforcement learning is able to utilize the increased information available from turbulence resolved simulations to learn improved, dynamic flow-responsive control for wind farm power
maximization, with direct implications for accelerating renewable energy deployment to net-zero targets
FEMA GRAS assessment of natural flavor complexes: vanilla extract, bitter almond oil, Wintergreen oil and related flavoring ingredients
No full textThe Expert Panel of the Flavor and Extract Manufacturers Association (FEMA) is conducting a program to re-evaluate the safety of over 250 natural flavor complexes (NFCs) used as flavoring ingredients. This publication, fourteenth in the series, evaluates the safety of NFCs composed primarily of benzaldehyde, methyl salicylate, vanillin and other benzyl derivative compounds. In 2018, the Expert Panel published an update of its safety evaluation procedure for NFCs that was first published in 2005. This procedure relies on a complete constituent characterization of the NFC and organization of the constituents of each NFC into defined congeneric groups. The safety of the NFC is evaluated using the threshold of toxicological concern (TTC) approach using updated estimates of exposure in addition to the evaluation of all relevant safety data on the NFC and its principal constituents. The scope of the safety evaluation contained herein does not include added use in dietary supplements or any products other than food. Eighteen (18) NFCs, derived from the Vanilla, Prunus, Betula, Acacia, Cuminum, Jasminum, Gaultheria, Polianthes and Evernia genera, were affirmed as generally recognized as safe (GRAS) under their conditions of intended use as flavor ingredients, based on an evaluation of each NFC and the constituents and congeneric groups therein
Advancements in face reconstruction from a single image
No full textThis Thesis addresses the fundamental challenge of 3D face reconstruction from a single image. Since the introduction of the seminal 3D Morphable Model (3DMM), significant advancements have been made in both human face representations, including implicit functions, and generation approaches, such as GANs and diffusion models. We explore the trade-off between these representations and generation techniques by proposing approaches that effectively leverage their strengths of both, for achieving high-fidelity results.
Firstly, we present 3DMM-RF, an implicit 3DMM that integrates a neural radiance field to a style-based GAN for face generation. By synthesizing the radiance field in one-pass, we overcome NeRF’s limitations of rigidity and slow rendering. Whilst trained on a large synthetic dataset, our model accurately reconstructs facial identities, under arbitrary pose, and appearance, enabling controllable re-rendering of “in-the-wild” images.
Next, we introduce FitDiff, a multi-modal latent diffusion model that generates relightable 3D facial avatars. It jointly outputs facial reflectance UV maps (diffuse albedo, specular albedo and normals) and shape given an input 2D image. During sampling, the proposed methodology achieves state-of-the-art 3D facial fitting as it utilizes a robust identity conditioning mechanism while employing perceptual and identity losses that guide the process.
Finally, SpinMeRound is a diffusion-based approach that operates directly in the image space, which can be used for synthesizing consistent all-around multi-view head portraits. Given an input facial image, it generates novel viewpoints of it, while preserving the crucial identity features. Our experiments show that SpinMeRound surpasses existing multi-view diffusion models in full-head synthesis.Open Acces
Participatory scenarios and spatial modelling to explore mangrove ecosystem services futures in Lamu, Kenya
No full textLand use and land cover change (LULCC) disrupts ecosystem structure and function, threatening ecosystem services and human well-being. Anticipating future trajectories is especially urgent in coastal regions, where mangrove ecosystems face anthropogenic and climatic pressures. In Lamu County, ongoing large-scale developments are expected to attract new settlements along the coastline and increase competition over land and natural resources, further intensifying pressures on mangroves. Here, we combine participatory scenario development with spatial modelling and ecosystem service valuation to explore plausible futures in Lamu County, Kenya. Using the Kesho, a diverse group of stakeholders co-produced four development scenarios to 2063, which were translated into spatially explicit LULCC maps using Landsat derived datasets and stakeholder-informed driver assumptions. A benefit transfer method was applied in two ways to estimate the value of mangroves’ provisioning, regulating and cultural services. Based on land-cover change alone, all scenarios show slight declines in value of ecosystem services. However, when scenario-specific changes in unit values were incorporated, the annual value diverged sharply, rising to USD 10.5 billion under the New Dawn scenario and falling to USD 7.6 billion under the Growth Trap. This study presents the first participatory scenario assessment in Lamu County, providing policy-relevant insights into how development pathways may shape mangrove ecosystems and the services they provide. Beyond Lamu, Kesho offers an adaptable tool for application in other mangrove-rich regions globally, supporting efforts to align local decision-making with continental and global sustainability agendas, including the African Union’s Agenda 2063 and the Sustainable Development Goals
Martian ionospheric response during the May 2024 solar superstorm
No full textSolar energetic events can have considerable effects on planetary ionospheres. However, the erratic nature of these solar energetic events make observations difficult. Here we show a mutual radio occultation observation, which serendipitously occurred just 10 minutes after a large solar flare impacted Mars. This resulted in the largest lower ionospheric layer ever recorded, where it was 278% its typical size. We used in-situ soft x-ray irradiance measurements to show a threefold increase in flux. This infers a different relation of soft X-ray to this layer’s density than previously thought, with variations depending on the amount of spectrum 'hardening' leading to the increase of ionisation from secondaries
Machine learning-driven nanopore sensing for quantitative, label-free miRNA detection
No full textNanopore sensors offer exceptional sensitivity for detecting single molecules, making them ideal for early disease diagnostics. In this study, we present a multiplexed nanopore-based assay that combines DNA-barcoded probes with advanced computational analysis to detect microRNAs (miRNAs) with high specificity and quantitative accuracy. Each probe selectively binds its target biomarker and induces a characteristic delay in the ionic current signal upon translocation through
the nanopore, enabling label-free detection.
We evaluated three analytical strategies for classifying delayed versus non-delayed events: (1) moving standard deviation (MSD), (2) spectral entropy (SE), and (3) a convolutional neural network (CNN). While MSD and SE rely on manually defined thresholds and exhibit limited
sensitivity, the CNN model, trained on image representations of raw current traces, achieved near-perfect classification performance across all metrics (accuracy = 0.99, precision = 0.99, recall = 0.99). Grad-CAM visualisation confirmed that the CNN focused on biophysically relevant signal regions, enhancing interpretability and generalisability. All methods produced sigmoidal concentration-response curves consistent with expected binding kinetics, and nanopore-derived delay metrics closely matched RT-qPCR validation data. All three methods were capable of distinguishing between signal classes; however, the CNN model demonstrated superior sensitivity and robustness. This work highlights the importance of data interpretation in nanopore sensing and presents a comparative framework for binary event classification. The findings pave the way for the development of machine learning-driven nanopore diagnostics capable of detecting diverse biomarker types at the single-molecule level
3D tibial HU reconstruction from biplanar X-rays utilizing a hybrid PCA-CNN framework
No full textHigh-resolution Computed Tomography (CT) is the gold standard medical imaging technique for bone assessment. However, its clinical use is limited by high radiation dose (8.8 mSv; biplanar X-rays 1.4 mSv), cost, and reduced accessibility. These barriers are particularly significant for patients requiring frequent imaging. This study introduces a novel hybrid framework combining statistical intensity modeling with Deep Learning to reconstruct 3D tibial CT volumes including internal density distributions from biplanar radiographs. The method employs principal component analysis (PCA) to capture intensity variations in a compact latent space and trains a convolutional neural network (CNN) to regress PCA coefficients directly from radiographs. The framework was developed and validated using 60 subjects from the publicly available Korea Institute of Science and Technology Information (KISTI) database. Compared to ground truth CT, it achieved a mean absolute error of 127.17 ± 12.08 Hounsfield Units (HU), a structural similarity index of 0.8558 ± 0.0215, and a peak signal-to-noise ratio of 21.40 ± 0.78 dB. The method has the potential to achieve substantial radiation dose reduction compared to conventional CT while preserving sufficient anatomical detail for potential clinical tasks such as patient-specific implant planning and bone quality triage. However, the actual dose reduction depends on clinical imaging protocols and requires validation through protocol-matched dosimetry on actual radiographs. Moreover, it produces interpretable outputs that reflect anatomical intensity variations (e.g., cortical vs. trabecular regions), demonstrating feasibility for hybrid statistical-Deep Learning bone reconstruction. The proposed pipeline establishes a foundation for reduced-dose 3D bone imaging and offers a pathway toward clinical translation pending validation on real-world radiographic data
Comparative outcomes of laparoscopic and robotic colorectal cancer surgery in the NHS: real-world evidence from sequential adoption of Versius and da Vinci Xi
No full textBackground Multiple robotic systems are now available for colorectal cancer surgery, yet comparative real-world evidence to guide NHS adoption remains limited. This study compared perioperative, oncological, and learning-curve outcomes for laparoscopic, Versius (CMR), and da Vinci Xi (dV) resections.
Methods A single-centre evaluation included elective colorectal cancer resections between November 2021 and May 2025 using laparoscopy, CMR or dV. Primary outcomes were length of stay (LOS) and operative time; secondary outcomes included lymph-node yield and Clavien–Dindo ≥2 complications. Analyses used non-parametric tests, Bonferroni-adjusted comparisons and multivariable regression. Learning curves were assessed with rolling means, LOWESS and CUSUM.
Results A total of 290 patients were included: laparoscopy (n=85), CMR (n=103), and dV (n=102). Median LOS was 5, 5 and 4 days respectively (p5 days) showed lower odds with dV versus CMR (OR 0.45, 95% CI 0.23–0.85; p=0.002), with adjusted probabilities of 22.1% for dV, 37.5% for CMR and 49.0% for laparoscopy. CMR had longer adjusted operative times than laparoscopy (+37.2 min; p=0.001). Lymph-node yield was highest with dV (median 25.5 vs 22; p=0.007), confirmed in adjusted analysis (+4.0 nodes; p=0.004). Major complications were similar; CMR rectal cases had
higher unadjusted rates. Learning curves showed operative time reduction for both robotic systems, with earlier plateauing for dV.
Conclusion Both robotic platforms were safe and oncologically equivalent; however, dV demonstrated shorter LOS, higher lymph node yield, and a more favourable learning curve
Data-driven modelling of N₂O production in wastewater processes using neural ordinary differential equations
No full textModelling nitrous oxide (N2O) production in wastewater treatment processes presents greater challenges than for other components, owing to its multiple production
pathways and pronounced spatiotemporal variations. This study proposes a novel data-driven approach employing neural ordinary differential equations (NODEs) to capture the intrinsic dynamics of N2O production in typical activated sludge processes. The NODE models are trained directly on state trajectory data, which incorporate continuous influent variations and operational adjustments as external forcings to the system dynamics. To address these external influences, we extend standard training procedures. In addition, a normalisation technique and an incremental strategy are
introduced to enhance the computational efficiency of NODE implementation in stiff wastewater systems. This methodology is validated using simulated data from the
benchmark simulation model no.1 (BSM1) plant, adapted to integrate the activated sludge model for greenhouse gases no.1 (ASMG1). Results demonstrate the efficacy of NODE-based approach in accurately capturing the complex dynamics governing N2O production, highlighting its potential for controlling and mitigating greenhouse gases emissions in wastewater treatment